We would not consider for a second today letting an AI just write Wikipedia articles because we know how bad they can be, Jimmy Wales added
By Sudhi Ranjan Sen and Dan Strumpf
The founder of Wikipedia said he isn’t worried about the threat posed to the free online encyclopedia from AI-generated content, including from Elon Musk’s Grokipedia, because of how error-prone the information tends to be.
“Why do I go to Wikipedia? I go to Wikipedia because it’s human-vetted knowledge,” said Jimmy Wales, the founder of the popular internet encyclopedia, whose articles are written and edited by human volunteers. “We would not consider for a second today letting an AI just write Wikipedia articles because we know how bad they can be. So I think that’s not really a concern.”
Among the problems with output from large language models like OpenAI’s ChatGPT and Alphabet Inc.’s Gemini, he said, is the high frequency with which they still generate “hallucinations,” or erroneous or misleading information.
It’s for that reason that he isn’t worried about competition from rivals such as Grokipedia, an AI-generated online encyclopedia launched last year by Musk’s xAI, which he called a “cartoon imitation of an encyclopedia.”
Wales made the comments in an interview this week on the sidelines of the India AI Impact Summit in New Delhi, an event that attracted more than a dozen heads of state and tech officials from OpenAI, Alphabet Inc., Anthropic PBC and others.
AI hallucinations become more flagrant and common as the subject matter becomes more obscure or niche, Wales said. The value of human-generated articles is that they benefit from contributions from subject-matter experts, which helps to guard against inaccuracies and makes for better-informed articles, he said.
“People are obsessives,” he said. “That sort of full, rich human context of understanding is actually quite important in terms of really understanding both what does the reader want and what does the reader need.”
A 2025 study by OpenAI found hallucinations were still common across even its advanced models, with hallucination rates as high as 79 per cent in some tests.
“The hallucination problem gets worse the more obscure the topic and therefore, for areas you might think we could use help, it’s actually very, very bad,” Wales said.
